Skip to main content

Table 1 Optimal number of latent classes: assessment statistics

From: Determining non-cigarette tobacco, alcohol, and substance use typologies across menthol and non-menthol smokers using latent class analysis

Model fit
Latent classes Number of free parameters LL BIC Sample size adjusted BIC LMR p-value VLMR LRT p-value Entropy
1 7 −4353.9 8757.3 8735.1 - - -
2 15 −4024.9 8155.9 8108.2 <0.0001 <0.0001 0.80
3 23 −3989.1 8140.9 8067.8 <0.0001 <0.0001 0.91
4 31 −3977.9 8175.0 8076.5 0.17 0.18 0.87
5 39 −3967.9 8211.6 8087.7 0.43 0.44 0.83
Odds of correct classification  
  Class 1 Class 2 Class 3 Class 4 Class 5   
1       
2 12.5 22.8      
3 12.0 8.1 54.6     
4 3.0 8.6 4.1 40.7    
5 3.2 4.1 11.2 4.0 24.0   
  1. Note. LL log likelihood, BIC Bayesian information criteria, LMR Lo-Mendell-Rubin, VLMR Vuong-Lo-Mendell-Rubin, LRT likelihood ratio test for k (H0) versus k-1 classes. Odds of correct classification (OCC) > 5 indicates a model with good latent class separation (Collins & Lanza, p. 74); OCC = ∞ indicates perfect classification
\